xref: /linux-6.15/include/linux/rcupdate.h (revision 8cb53b48)
1 /* SPDX-License-Identifier: GPL-2.0+ */
2 /*
3  * Read-Copy Update mechanism for mutual exclusion
4  *
5  * Copyright IBM Corporation, 2001
6  *
7  * Author: Dipankar Sarma <[email protected]>
8  *
9  * Based on the original work by Paul McKenney <[email protected]>
10  * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
11  * Papers:
12  * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
13  * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
14  *
15  * For detailed explanation of Read-Copy Update mechanism see -
16  *		http://lse.sourceforge.net/locking/rcupdate.html
17  *
18  */
19 
20 #ifndef __LINUX_RCUPDATE_H
21 #define __LINUX_RCUPDATE_H
22 
23 #include <linux/types.h>
24 #include <linux/compiler.h>
25 #include <linux/atomic.h>
26 #include <linux/irqflags.h>
27 #include <linux/preempt.h>
28 #include <linux/bottom_half.h>
29 #include <linux/lockdep.h>
30 #include <asm/processor.h>
31 #include <linux/cpumask.h>
32 #include <linux/context_tracking_irq.h>
33 
34 #define ULONG_CMP_GE(a, b)	(ULONG_MAX / 2 >= (a) - (b))
35 #define ULONG_CMP_LT(a, b)	(ULONG_MAX / 2 < (a) - (b))
36 #define ulong2long(a)		(*(long *)(&(a)))
37 #define USHORT_CMP_GE(a, b)	(USHRT_MAX / 2 >= (unsigned short)((a) - (b)))
38 #define USHORT_CMP_LT(a, b)	(USHRT_MAX / 2 < (unsigned short)((a) - (b)))
39 
40 /* Exported common interfaces */
41 void call_rcu(struct rcu_head *head, rcu_callback_t func);
42 void rcu_barrier_tasks(void);
43 void rcu_barrier_tasks_rude(void);
44 void synchronize_rcu(void);
45 
46 struct rcu_gp_oldstate;
47 unsigned long get_completed_synchronize_rcu(void);
48 void get_completed_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp);
49 
50 // Maximum number of unsigned long values corresponding to
51 // not-yet-completed RCU grace periods.
52 #define NUM_ACTIVE_RCU_POLL_OLDSTATE 2
53 
54 /**
55  * same_state_synchronize_rcu - Are two old-state values identical?
56  * @oldstate1: First old-state value.
57  * @oldstate2: Second old-state value.
58  *
59  * The two old-state values must have been obtained from either
60  * get_state_synchronize_rcu(), start_poll_synchronize_rcu(), or
61  * get_completed_synchronize_rcu().  Returns @true if the two values are
62  * identical and @false otherwise.  This allows structures whose lifetimes
63  * are tracked by old-state values to push these values to a list header,
64  * allowing those structures to be slightly smaller.
65  */
66 static inline bool same_state_synchronize_rcu(unsigned long oldstate1, unsigned long oldstate2)
67 {
68 	return oldstate1 == oldstate2;
69 }
70 
71 #ifdef CONFIG_PREEMPT_RCU
72 
73 void __rcu_read_lock(void);
74 void __rcu_read_unlock(void);
75 
76 /*
77  * Defined as a macro as it is a very low level header included from
78  * areas that don't even know about current.  This gives the rcu_read_lock()
79  * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
80  * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
81  */
82 #define rcu_preempt_depth() READ_ONCE(current->rcu_read_lock_nesting)
83 
84 #else /* #ifdef CONFIG_PREEMPT_RCU */
85 
86 #ifdef CONFIG_TINY_RCU
87 #define rcu_read_unlock_strict() do { } while (0)
88 #else
89 void rcu_read_unlock_strict(void);
90 #endif
91 
92 static inline void __rcu_read_lock(void)
93 {
94 	preempt_disable();
95 }
96 
97 static inline void __rcu_read_unlock(void)
98 {
99 	preempt_enable();
100 	if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD))
101 		rcu_read_unlock_strict();
102 }
103 
104 static inline int rcu_preempt_depth(void)
105 {
106 	return 0;
107 }
108 
109 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
110 
111 /* Internal to kernel */
112 void rcu_init(void);
113 extern int rcu_scheduler_active;
114 void rcu_sched_clock_irq(int user);
115 void rcu_report_dead(unsigned int cpu);
116 void rcutree_migrate_callbacks(int cpu);
117 
118 #ifdef CONFIG_TASKS_RCU_GENERIC
119 void rcu_init_tasks_generic(void);
120 #else
121 static inline void rcu_init_tasks_generic(void) { }
122 #endif
123 
124 #ifdef CONFIG_RCU_STALL_COMMON
125 void rcu_sysrq_start(void);
126 void rcu_sysrq_end(void);
127 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
128 static inline void rcu_sysrq_start(void) { }
129 static inline void rcu_sysrq_end(void) { }
130 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
131 
132 #if defined(CONFIG_NO_HZ_FULL) && (!defined(CONFIG_GENERIC_ENTRY) || !defined(CONFIG_KVM_XFER_TO_GUEST_WORK))
133 void rcu_irq_work_resched(void);
134 #else
135 static inline void rcu_irq_work_resched(void) { }
136 #endif
137 
138 #ifdef CONFIG_RCU_NOCB_CPU
139 void rcu_init_nohz(void);
140 int rcu_nocb_cpu_offload(int cpu);
141 int rcu_nocb_cpu_deoffload(int cpu);
142 void rcu_nocb_flush_deferred_wakeup(void);
143 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
144 static inline void rcu_init_nohz(void) { }
145 static inline int rcu_nocb_cpu_offload(int cpu) { return -EINVAL; }
146 static inline int rcu_nocb_cpu_deoffload(int cpu) { return 0; }
147 static inline void rcu_nocb_flush_deferred_wakeup(void) { }
148 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
149 
150 /**
151  * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
152  * @a: Code that RCU needs to pay attention to.
153  *
154  * RCU read-side critical sections are forbidden in the inner idle loop,
155  * that is, between the ct_idle_enter() and the ct_idle_exit() -- RCU
156  * will happily ignore any such read-side critical sections.  However,
157  * things like powertop need tracepoints in the inner idle loop.
158  *
159  * This macro provides the way out:  RCU_NONIDLE(do_something_with_RCU())
160  * will tell RCU that it needs to pay attention, invoke its argument
161  * (in this example, calling the do_something_with_RCU() function),
162  * and then tell RCU to go back to ignoring this CPU.  It is permissible
163  * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is
164  * on the order of a million or so, even on 32-bit systems).  It is
165  * not legal to block within RCU_NONIDLE(), nor is it permissible to
166  * transfer control either into or out of RCU_NONIDLE()'s statement.
167  */
168 #define RCU_NONIDLE(a) \
169 	do { \
170 		ct_irq_enter_irqson(); \
171 		do { a; } while (0); \
172 		ct_irq_exit_irqson(); \
173 	} while (0)
174 
175 /*
176  * Note a quasi-voluntary context switch for RCU-tasks's benefit.
177  * This is a macro rather than an inline function to avoid #include hell.
178  */
179 #ifdef CONFIG_TASKS_RCU_GENERIC
180 
181 # ifdef CONFIG_TASKS_RCU
182 # define rcu_tasks_classic_qs(t, preempt)				\
183 	do {								\
184 		if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout))	\
185 			WRITE_ONCE((t)->rcu_tasks_holdout, false);	\
186 	} while (0)
187 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func);
188 void synchronize_rcu_tasks(void);
189 # else
190 # define rcu_tasks_classic_qs(t, preempt) do { } while (0)
191 # define call_rcu_tasks call_rcu
192 # define synchronize_rcu_tasks synchronize_rcu
193 # endif
194 
195 # ifdef CONFIG_TASKS_TRACE_RCU
196 // Bits for ->trc_reader_special.b.need_qs field.
197 #define TRC_NEED_QS		0x1  // Task needs a quiescent state.
198 #define TRC_NEED_QS_CHECKED	0x2  // Task has been checked for needing quiescent state.
199 
200 u8 rcu_trc_cmpxchg_need_qs(struct task_struct *t, u8 old, u8 new);
201 void rcu_tasks_trace_qs_blkd(struct task_struct *t);
202 
203 # define rcu_tasks_trace_qs(t)							\
204 	do {									\
205 		int ___rttq_nesting = READ_ONCE((t)->trc_reader_nesting);	\
206 										\
207 		if (likely(!READ_ONCE((t)->trc_reader_special.b.need_qs)) &&	\
208 		    likely(!___rttq_nesting)) {					\
209 			rcu_trc_cmpxchg_need_qs((t), 0,	TRC_NEED_QS_CHECKED);	\
210 		} else if (___rttq_nesting && ___rttq_nesting != INT_MIN &&	\
211 			   !READ_ONCE((t)->trc_reader_special.b.blocked)) {	\
212 			rcu_tasks_trace_qs_blkd(t);				\
213 		}								\
214 	} while (0)
215 # else
216 # define rcu_tasks_trace_qs(t) do { } while (0)
217 # endif
218 
219 #define rcu_tasks_qs(t, preempt)					\
220 do {									\
221 	rcu_tasks_classic_qs((t), (preempt));				\
222 	rcu_tasks_trace_qs(t);						\
223 } while (0)
224 
225 # ifdef CONFIG_TASKS_RUDE_RCU
226 void call_rcu_tasks_rude(struct rcu_head *head, rcu_callback_t func);
227 void synchronize_rcu_tasks_rude(void);
228 # endif
229 
230 #define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false)
231 void exit_tasks_rcu_start(void);
232 void exit_tasks_rcu_finish(void);
233 #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */
234 #define rcu_tasks_classic_qs(t, preempt) do { } while (0)
235 #define rcu_tasks_qs(t, preempt) do { } while (0)
236 #define rcu_note_voluntary_context_switch(t) do { } while (0)
237 #define call_rcu_tasks call_rcu
238 #define synchronize_rcu_tasks synchronize_rcu
239 static inline void exit_tasks_rcu_start(void) { }
240 static inline void exit_tasks_rcu_finish(void) { }
241 #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */
242 
243 /**
244  * rcu_trace_implies_rcu_gp - does an RCU Tasks Trace grace period imply an RCU grace period?
245  *
246  * As an accident of implementation, an RCU Tasks Trace grace period also
247  * acts as an RCU grace period.  However, this could change at any time.
248  * Code relying on this accident must call this function to verify that
249  * this accident is still happening.
250  *
251  * You have been warned!
252  */
253 static inline bool rcu_trace_implies_rcu_gp(void) { return true; }
254 
255 /**
256  * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU
257  *
258  * This macro resembles cond_resched(), except that it is defined to
259  * report potential quiescent states to RCU-tasks even if the cond_resched()
260  * machinery were to be shut off, as some advocate for PREEMPTION kernels.
261  */
262 #define cond_resched_tasks_rcu_qs() \
263 do { \
264 	rcu_tasks_qs(current, false); \
265 	cond_resched(); \
266 } while (0)
267 
268 /*
269  * Infrastructure to implement the synchronize_() primitives in
270  * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
271  */
272 
273 #if defined(CONFIG_TREE_RCU)
274 #include <linux/rcutree.h>
275 #elif defined(CONFIG_TINY_RCU)
276 #include <linux/rcutiny.h>
277 #else
278 #error "Unknown RCU implementation specified to kernel configuration"
279 #endif
280 
281 /*
282  * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls
283  * are needed for dynamic initialization and destruction of rcu_head
284  * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for
285  * dynamic initialization and destruction of statically allocated rcu_head
286  * structures.  However, rcu_head structures allocated dynamically in the
287  * heap don't need any initialization.
288  */
289 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
290 void init_rcu_head(struct rcu_head *head);
291 void destroy_rcu_head(struct rcu_head *head);
292 void init_rcu_head_on_stack(struct rcu_head *head);
293 void destroy_rcu_head_on_stack(struct rcu_head *head);
294 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
295 static inline void init_rcu_head(struct rcu_head *head) { }
296 static inline void destroy_rcu_head(struct rcu_head *head) { }
297 static inline void init_rcu_head_on_stack(struct rcu_head *head) { }
298 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { }
299 #endif	/* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
300 
301 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
302 bool rcu_lockdep_current_cpu_online(void);
303 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
304 static inline bool rcu_lockdep_current_cpu_online(void) { return true; }
305 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
306 
307 extern struct lockdep_map rcu_lock_map;
308 extern struct lockdep_map rcu_bh_lock_map;
309 extern struct lockdep_map rcu_sched_lock_map;
310 extern struct lockdep_map rcu_callback_map;
311 
312 #ifdef CONFIG_DEBUG_LOCK_ALLOC
313 
314 static inline void rcu_lock_acquire(struct lockdep_map *map)
315 {
316 	lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
317 }
318 
319 static inline void rcu_lock_release(struct lockdep_map *map)
320 {
321 	lock_release(map, _THIS_IP_);
322 }
323 
324 int debug_lockdep_rcu_enabled(void);
325 int rcu_read_lock_held(void);
326 int rcu_read_lock_bh_held(void);
327 int rcu_read_lock_sched_held(void);
328 int rcu_read_lock_any_held(void);
329 
330 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
331 
332 # define rcu_lock_acquire(a)		do { } while (0)
333 # define rcu_lock_release(a)		do { } while (0)
334 
335 static inline int rcu_read_lock_held(void)
336 {
337 	return 1;
338 }
339 
340 static inline int rcu_read_lock_bh_held(void)
341 {
342 	return 1;
343 }
344 
345 static inline int rcu_read_lock_sched_held(void)
346 {
347 	return !preemptible();
348 }
349 
350 static inline int rcu_read_lock_any_held(void)
351 {
352 	return !preemptible();
353 }
354 
355 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
356 
357 #ifdef CONFIG_PROVE_RCU
358 
359 /**
360  * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
361  * @c: condition to check
362  * @s: informative message
363  */
364 #define RCU_LOCKDEP_WARN(c, s)						\
365 	do {								\
366 		static bool __section(".data.unlikely") __warned;	\
367 		if ((c) && debug_lockdep_rcu_enabled() && !__warned) {	\
368 			__warned = true;				\
369 			lockdep_rcu_suspicious(__FILE__, __LINE__, s);	\
370 		}							\
371 	} while (0)
372 
373 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
374 static inline void rcu_preempt_sleep_check(void)
375 {
376 	RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map),
377 			 "Illegal context switch in RCU read-side critical section");
378 }
379 #else /* #ifdef CONFIG_PROVE_RCU */
380 static inline void rcu_preempt_sleep_check(void) { }
381 #endif /* #else #ifdef CONFIG_PROVE_RCU */
382 
383 #define rcu_sleep_check()						\
384 	do {								\
385 		rcu_preempt_sleep_check();				\
386 		if (!IS_ENABLED(CONFIG_PREEMPT_RT))			\
387 		    RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map),	\
388 				 "Illegal context switch in RCU-bh read-side critical section"); \
389 		RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map),	\
390 				 "Illegal context switch in RCU-sched read-side critical section"); \
391 	} while (0)
392 
393 #else /* #ifdef CONFIG_PROVE_RCU */
394 
395 #define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c))
396 #define rcu_sleep_check() do { } while (0)
397 
398 #endif /* #else #ifdef CONFIG_PROVE_RCU */
399 
400 /*
401  * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
402  * and rcu_assign_pointer().  Some of these could be folded into their
403  * callers, but they are left separate in order to ease introduction of
404  * multiple pointers markings to match different RCU implementations
405  * (e.g., __srcu), should this make sense in the future.
406  */
407 
408 #ifdef __CHECKER__
409 #define rcu_check_sparse(p, space) \
410 	((void)(((typeof(*p) space *)p) == p))
411 #else /* #ifdef __CHECKER__ */
412 #define rcu_check_sparse(p, space)
413 #endif /* #else #ifdef __CHECKER__ */
414 
415 #define __unrcu_pointer(p, local)					\
416 ({									\
417 	typeof(*p) *local = (typeof(*p) *__force)(p);			\
418 	rcu_check_sparse(p, __rcu);					\
419 	((typeof(*p) __force __kernel *)(local)); 			\
420 })
421 /**
422  * unrcu_pointer - mark a pointer as not being RCU protected
423  * @p: pointer needing to lose its __rcu property
424  *
425  * Converts @p from an __rcu pointer to a __kernel pointer.
426  * This allows an __rcu pointer to be used with xchg() and friends.
427  */
428 #define unrcu_pointer(p) __unrcu_pointer(p, __UNIQUE_ID(rcu))
429 
430 #define __rcu_access_pointer(p, local, space) \
431 ({ \
432 	typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \
433 	rcu_check_sparse(p, space); \
434 	((typeof(*p) __force __kernel *)(local)); \
435 })
436 #define __rcu_dereference_check(p, local, c, space) \
437 ({ \
438 	/* Dependency order vs. p above. */ \
439 	typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \
440 	RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
441 	rcu_check_sparse(p, space); \
442 	((typeof(*p) __force __kernel *)(local)); \
443 })
444 #define __rcu_dereference_protected(p, local, c, space) \
445 ({ \
446 	RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
447 	rcu_check_sparse(p, space); \
448 	((typeof(*p) __force __kernel *)(p)); \
449 })
450 #define __rcu_dereference_raw(p, local) \
451 ({ \
452 	/* Dependency order vs. p above. */ \
453 	typeof(p) local = READ_ONCE(p); \
454 	((typeof(*p) __force __kernel *)(local)); \
455 })
456 #define rcu_dereference_raw(p) __rcu_dereference_raw(p, __UNIQUE_ID(rcu))
457 
458 /**
459  * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
460  * @v: The value to statically initialize with.
461  */
462 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
463 
464 /**
465  * rcu_assign_pointer() - assign to RCU-protected pointer
466  * @p: pointer to assign to
467  * @v: value to assign (publish)
468  *
469  * Assigns the specified value to the specified RCU-protected
470  * pointer, ensuring that any concurrent RCU readers will see
471  * any prior initialization.
472  *
473  * Inserts memory barriers on architectures that require them
474  * (which is most of them), and also prevents the compiler from
475  * reordering the code that initializes the structure after the pointer
476  * assignment.  More importantly, this call documents which pointers
477  * will be dereferenced by RCU read-side code.
478  *
479  * In some special cases, you may use RCU_INIT_POINTER() instead
480  * of rcu_assign_pointer().  RCU_INIT_POINTER() is a bit faster due
481  * to the fact that it does not constrain either the CPU or the compiler.
482  * That said, using RCU_INIT_POINTER() when you should have used
483  * rcu_assign_pointer() is a very bad thing that results in
484  * impossible-to-diagnose memory corruption.  So please be careful.
485  * See the RCU_INIT_POINTER() comment header for details.
486  *
487  * Note that rcu_assign_pointer() evaluates each of its arguments only
488  * once, appearances notwithstanding.  One of the "extra" evaluations
489  * is in typeof() and the other visible only to sparse (__CHECKER__),
490  * neither of which actually execute the argument.  As with most cpp
491  * macros, this execute-arguments-only-once property is important, so
492  * please be careful when making changes to rcu_assign_pointer() and the
493  * other macros that it invokes.
494  */
495 #define rcu_assign_pointer(p, v)					      \
496 do {									      \
497 	uintptr_t _r_a_p__v = (uintptr_t)(v);				      \
498 	rcu_check_sparse(p, __rcu);					      \
499 									      \
500 	if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL)	      \
501 		WRITE_ONCE((p), (typeof(p))(_r_a_p__v));		      \
502 	else								      \
503 		smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
504 } while (0)
505 
506 /**
507  * rcu_replace_pointer() - replace an RCU pointer, returning its old value
508  * @rcu_ptr: RCU pointer, whose old value is returned
509  * @ptr: regular pointer
510  * @c: the lockdep conditions under which the dereference will take place
511  *
512  * Perform a replacement, where @rcu_ptr is an RCU-annotated
513  * pointer and @c is the lockdep argument that is passed to the
514  * rcu_dereference_protected() call used to read that pointer.  The old
515  * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr.
516  */
517 #define rcu_replace_pointer(rcu_ptr, ptr, c)				\
518 ({									\
519 	typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c));	\
520 	rcu_assign_pointer((rcu_ptr), (ptr));				\
521 	__tmp;								\
522 })
523 
524 /**
525  * rcu_access_pointer() - fetch RCU pointer with no dereferencing
526  * @p: The pointer to read
527  *
528  * Return the value of the specified RCU-protected pointer, but omit the
529  * lockdep checks for being in an RCU read-side critical section.  This is
530  * useful when the value of this pointer is accessed, but the pointer is
531  * not dereferenced, for example, when testing an RCU-protected pointer
532  * against NULL.  Although rcu_access_pointer() may also be used in cases
533  * where update-side locks prevent the value of the pointer from changing,
534  * you should instead use rcu_dereference_protected() for this use case.
535  * Within an RCU read-side critical section, there is little reason to
536  * use rcu_access_pointer().
537  *
538  * It is usually best to test the rcu_access_pointer() return value
539  * directly in order to avoid accidental dereferences being introduced
540  * by later inattentive changes.  In other words, assigning the
541  * rcu_access_pointer() return value to a local variable results in an
542  * accident waiting to happen.
543  *
544  * It is also permissible to use rcu_access_pointer() when read-side
545  * access to the pointer was removed at least one grace period ago, as is
546  * the case in the context of the RCU callback that is freeing up the data,
547  * or after a synchronize_rcu() returns.  This can be useful when tearing
548  * down multi-linked structures after a grace period has elapsed.  However,
549  * rcu_dereference_protected() is normally preferred for this use case.
550  */
551 #define rcu_access_pointer(p) __rcu_access_pointer((p), __UNIQUE_ID(rcu), __rcu)
552 
553 /**
554  * rcu_dereference_check() - rcu_dereference with debug checking
555  * @p: The pointer to read, prior to dereferencing
556  * @c: The conditions under which the dereference will take place
557  *
558  * Do an rcu_dereference(), but check that the conditions under which the
559  * dereference will take place are correct.  Typically the conditions
560  * indicate the various locking conditions that should be held at that
561  * point.  The check should return true if the conditions are satisfied.
562  * An implicit check for being in an RCU read-side critical section
563  * (rcu_read_lock()) is included.
564  *
565  * For example:
566  *
567  *	bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
568  *
569  * could be used to indicate to lockdep that foo->bar may only be dereferenced
570  * if either rcu_read_lock() is held, or that the lock required to replace
571  * the bar struct at foo->bar is held.
572  *
573  * Note that the list of conditions may also include indications of when a lock
574  * need not be held, for example during initialisation or destruction of the
575  * target struct:
576  *
577  *	bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
578  *					      atomic_read(&foo->usage) == 0);
579  *
580  * Inserts memory barriers on architectures that require them
581  * (currently only the Alpha), prevents the compiler from refetching
582  * (and from merging fetches), and, more importantly, documents exactly
583  * which pointers are protected by RCU and checks that the pointer is
584  * annotated as __rcu.
585  */
586 #define rcu_dereference_check(p, c) \
587 	__rcu_dereference_check((p), __UNIQUE_ID(rcu), \
588 				(c) || rcu_read_lock_held(), __rcu)
589 
590 /**
591  * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
592  * @p: The pointer to read, prior to dereferencing
593  * @c: The conditions under which the dereference will take place
594  *
595  * This is the RCU-bh counterpart to rcu_dereference_check().  However,
596  * please note that starting in v5.0 kernels, vanilla RCU grace periods
597  * wait for local_bh_disable() regions of code in addition to regions of
598  * code demarked by rcu_read_lock() and rcu_read_unlock().  This means
599  * that synchronize_rcu(), call_rcu, and friends all take not only
600  * rcu_read_lock() but also rcu_read_lock_bh() into account.
601  */
602 #define rcu_dereference_bh_check(p, c) \
603 	__rcu_dereference_check((p), __UNIQUE_ID(rcu), \
604 				(c) || rcu_read_lock_bh_held(), __rcu)
605 
606 /**
607  * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
608  * @p: The pointer to read, prior to dereferencing
609  * @c: The conditions under which the dereference will take place
610  *
611  * This is the RCU-sched counterpart to rcu_dereference_check().
612  * However, please note that starting in v5.0 kernels, vanilla RCU grace
613  * periods wait for preempt_disable() regions of code in addition to
614  * regions of code demarked by rcu_read_lock() and rcu_read_unlock().
615  * This means that synchronize_rcu(), call_rcu, and friends all take not
616  * only rcu_read_lock() but also rcu_read_lock_sched() into account.
617  */
618 #define rcu_dereference_sched_check(p, c) \
619 	__rcu_dereference_check((p), __UNIQUE_ID(rcu), \
620 				(c) || rcu_read_lock_sched_held(), \
621 				__rcu)
622 
623 /*
624  * The tracing infrastructure traces RCU (we want that), but unfortunately
625  * some of the RCU checks causes tracing to lock up the system.
626  *
627  * The no-tracing version of rcu_dereference_raw() must not call
628  * rcu_read_lock_held().
629  */
630 #define rcu_dereference_raw_check(p) \
631 	__rcu_dereference_check((p), __UNIQUE_ID(rcu), 1, __rcu)
632 
633 /**
634  * rcu_dereference_protected() - fetch RCU pointer when updates prevented
635  * @p: The pointer to read, prior to dereferencing
636  * @c: The conditions under which the dereference will take place
637  *
638  * Return the value of the specified RCU-protected pointer, but omit
639  * the READ_ONCE().  This is useful in cases where update-side locks
640  * prevent the value of the pointer from changing.  Please note that this
641  * primitive does *not* prevent the compiler from repeating this reference
642  * or combining it with other references, so it should not be used without
643  * protection of appropriate locks.
644  *
645  * This function is only for update-side use.  Using this function
646  * when protected only by rcu_read_lock() will result in infrequent
647  * but very ugly failures.
648  */
649 #define rcu_dereference_protected(p, c) \
650 	__rcu_dereference_protected((p), __UNIQUE_ID(rcu), (c), __rcu)
651 
652 
653 /**
654  * rcu_dereference() - fetch RCU-protected pointer for dereferencing
655  * @p: The pointer to read, prior to dereferencing
656  *
657  * This is a simple wrapper around rcu_dereference_check().
658  */
659 #define rcu_dereference(p) rcu_dereference_check(p, 0)
660 
661 /**
662  * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
663  * @p: The pointer to read, prior to dereferencing
664  *
665  * Makes rcu_dereference_check() do the dirty work.
666  */
667 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
668 
669 /**
670  * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
671  * @p: The pointer to read, prior to dereferencing
672  *
673  * Makes rcu_dereference_check() do the dirty work.
674  */
675 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
676 
677 /**
678  * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
679  * @p: The pointer to hand off
680  *
681  * This is simply an identity function, but it documents where a pointer
682  * is handed off from RCU to some other synchronization mechanism, for
683  * example, reference counting or locking.  In C11, it would map to
684  * kill_dependency().  It could be used as follows::
685  *
686  *	rcu_read_lock();
687  *	p = rcu_dereference(gp);
688  *	long_lived = is_long_lived(p);
689  *	if (long_lived) {
690  *		if (!atomic_inc_not_zero(p->refcnt))
691  *			long_lived = false;
692  *		else
693  *			p = rcu_pointer_handoff(p);
694  *	}
695  *	rcu_read_unlock();
696  */
697 #define rcu_pointer_handoff(p) (p)
698 
699 /**
700  * rcu_read_lock() - mark the beginning of an RCU read-side critical section
701  *
702  * When synchronize_rcu() is invoked on one CPU while other CPUs
703  * are within RCU read-side critical sections, then the
704  * synchronize_rcu() is guaranteed to block until after all the other
705  * CPUs exit their critical sections.  Similarly, if call_rcu() is invoked
706  * on one CPU while other CPUs are within RCU read-side critical
707  * sections, invocation of the corresponding RCU callback is deferred
708  * until after the all the other CPUs exit their critical sections.
709  *
710  * In v5.0 and later kernels, synchronize_rcu() and call_rcu() also
711  * wait for regions of code with preemption disabled, including regions of
712  * code with interrupts or softirqs disabled.  In pre-v5.0 kernels, which
713  * define synchronize_sched(), only code enclosed within rcu_read_lock()
714  * and rcu_read_unlock() are guaranteed to be waited for.
715  *
716  * Note, however, that RCU callbacks are permitted to run concurrently
717  * with new RCU read-side critical sections.  One way that this can happen
718  * is via the following sequence of events: (1) CPU 0 enters an RCU
719  * read-side critical section, (2) CPU 1 invokes call_rcu() to register
720  * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
721  * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
722  * callback is invoked.  This is legal, because the RCU read-side critical
723  * section that was running concurrently with the call_rcu() (and which
724  * therefore might be referencing something that the corresponding RCU
725  * callback would free up) has completed before the corresponding
726  * RCU callback is invoked.
727  *
728  * RCU read-side critical sections may be nested.  Any deferred actions
729  * will be deferred until the outermost RCU read-side critical section
730  * completes.
731  *
732  * You can avoid reading and understanding the next paragraph by
733  * following this rule: don't put anything in an rcu_read_lock() RCU
734  * read-side critical section that would block in a !PREEMPTION kernel.
735  * But if you want the full story, read on!
736  *
737  * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU),
738  * it is illegal to block while in an RCU read-side critical section.
739  * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION
740  * kernel builds, RCU read-side critical sections may be preempted,
741  * but explicit blocking is illegal.  Finally, in preemptible RCU
742  * implementations in real-time (with -rt patchset) kernel builds, RCU
743  * read-side critical sections may be preempted and they may also block, but
744  * only when acquiring spinlocks that are subject to priority inheritance.
745  */
746 static __always_inline void rcu_read_lock(void)
747 {
748 	__rcu_read_lock();
749 	__acquire(RCU);
750 	rcu_lock_acquire(&rcu_lock_map);
751 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
752 			 "rcu_read_lock() used illegally while idle");
753 }
754 
755 /*
756  * So where is rcu_write_lock()?  It does not exist, as there is no
757  * way for writers to lock out RCU readers.  This is a feature, not
758  * a bug -- this property is what provides RCU's performance benefits.
759  * Of course, writers must coordinate with each other.  The normal
760  * spinlock primitives work well for this, but any other technique may be
761  * used as well.  RCU does not care how the writers keep out of each
762  * others' way, as long as they do so.
763  */
764 
765 /**
766  * rcu_read_unlock() - marks the end of an RCU read-side critical section.
767  *
768  * In almost all situations, rcu_read_unlock() is immune from deadlock.
769  * In recent kernels that have consolidated synchronize_sched() and
770  * synchronize_rcu_bh() into synchronize_rcu(), this deadlock immunity
771  * also extends to the scheduler's runqueue and priority-inheritance
772  * spinlocks, courtesy of the quiescent-state deferral that is carried
773  * out when rcu_read_unlock() is invoked with interrupts disabled.
774  *
775  * See rcu_read_lock() for more information.
776  */
777 static inline void rcu_read_unlock(void)
778 {
779 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
780 			 "rcu_read_unlock() used illegally while idle");
781 	__release(RCU);
782 	__rcu_read_unlock();
783 	rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
784 }
785 
786 /**
787  * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
788  *
789  * This is equivalent to rcu_read_lock(), but also disables softirqs.
790  * Note that anything else that disables softirqs can also serve as an RCU
791  * read-side critical section.  However, please note that this equivalence
792  * applies only to v5.0 and later.  Before v5.0, rcu_read_lock() and
793  * rcu_read_lock_bh() were unrelated.
794  *
795  * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
796  * must occur in the same context, for example, it is illegal to invoke
797  * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
798  * was invoked from some other task.
799  */
800 static inline void rcu_read_lock_bh(void)
801 {
802 	local_bh_disable();
803 	__acquire(RCU_BH);
804 	rcu_lock_acquire(&rcu_bh_lock_map);
805 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
806 			 "rcu_read_lock_bh() used illegally while idle");
807 }
808 
809 /**
810  * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section
811  *
812  * See rcu_read_lock_bh() for more information.
813  */
814 static inline void rcu_read_unlock_bh(void)
815 {
816 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
817 			 "rcu_read_unlock_bh() used illegally while idle");
818 	rcu_lock_release(&rcu_bh_lock_map);
819 	__release(RCU_BH);
820 	local_bh_enable();
821 }
822 
823 /**
824  * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
825  *
826  * This is equivalent to rcu_read_lock(), but also disables preemption.
827  * Read-side critical sections can also be introduced by anything else that
828  * disables preemption, including local_irq_disable() and friends.  However,
829  * please note that the equivalence to rcu_read_lock() applies only to
830  * v5.0 and later.  Before v5.0, rcu_read_lock() and rcu_read_lock_sched()
831  * were unrelated.
832  *
833  * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
834  * must occur in the same context, for example, it is illegal to invoke
835  * rcu_read_unlock_sched() from process context if the matching
836  * rcu_read_lock_sched() was invoked from an NMI handler.
837  */
838 static inline void rcu_read_lock_sched(void)
839 {
840 	preempt_disable();
841 	__acquire(RCU_SCHED);
842 	rcu_lock_acquire(&rcu_sched_lock_map);
843 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
844 			 "rcu_read_lock_sched() used illegally while idle");
845 }
846 
847 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
848 static inline notrace void rcu_read_lock_sched_notrace(void)
849 {
850 	preempt_disable_notrace();
851 	__acquire(RCU_SCHED);
852 }
853 
854 /**
855  * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section
856  *
857  * See rcu_read_lock_sched() for more information.
858  */
859 static inline void rcu_read_unlock_sched(void)
860 {
861 	RCU_LOCKDEP_WARN(!rcu_is_watching(),
862 			 "rcu_read_unlock_sched() used illegally while idle");
863 	rcu_lock_release(&rcu_sched_lock_map);
864 	__release(RCU_SCHED);
865 	preempt_enable();
866 }
867 
868 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
869 static inline notrace void rcu_read_unlock_sched_notrace(void)
870 {
871 	__release(RCU_SCHED);
872 	preempt_enable_notrace();
873 }
874 
875 /**
876  * RCU_INIT_POINTER() - initialize an RCU protected pointer
877  * @p: The pointer to be initialized.
878  * @v: The value to initialized the pointer to.
879  *
880  * Initialize an RCU-protected pointer in special cases where readers
881  * do not need ordering constraints on the CPU or the compiler.  These
882  * special cases are:
883  *
884  * 1.	This use of RCU_INIT_POINTER() is NULLing out the pointer *or*
885  * 2.	The caller has taken whatever steps are required to prevent
886  *	RCU readers from concurrently accessing this pointer *or*
887  * 3.	The referenced data structure has already been exposed to
888  *	readers either at compile time or via rcu_assign_pointer() *and*
889  *
890  *	a.	You have not made *any* reader-visible changes to
891  *		this structure since then *or*
892  *	b.	It is OK for readers accessing this structure from its
893  *		new location to see the old state of the structure.  (For
894  *		example, the changes were to statistical counters or to
895  *		other state where exact synchronization is not required.)
896  *
897  * Failure to follow these rules governing use of RCU_INIT_POINTER() will
898  * result in impossible-to-diagnose memory corruption.  As in the structures
899  * will look OK in crash dumps, but any concurrent RCU readers might
900  * see pre-initialized values of the referenced data structure.  So
901  * please be very careful how you use RCU_INIT_POINTER()!!!
902  *
903  * If you are creating an RCU-protected linked structure that is accessed
904  * by a single external-to-structure RCU-protected pointer, then you may
905  * use RCU_INIT_POINTER() to initialize the internal RCU-protected
906  * pointers, but you must use rcu_assign_pointer() to initialize the
907  * external-to-structure pointer *after* you have completely initialized
908  * the reader-accessible portions of the linked structure.
909  *
910  * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
911  * ordering guarantees for either the CPU or the compiler.
912  */
913 #define RCU_INIT_POINTER(p, v) \
914 	do { \
915 		rcu_check_sparse(p, __rcu); \
916 		WRITE_ONCE(p, RCU_INITIALIZER(v)); \
917 	} while (0)
918 
919 /**
920  * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
921  * @p: The pointer to be initialized.
922  * @v: The value to initialized the pointer to.
923  *
924  * GCC-style initialization for an RCU-protected pointer in a structure field.
925  */
926 #define RCU_POINTER_INITIALIZER(p, v) \
927 		.p = RCU_INITIALIZER(v)
928 
929 /*
930  * Does the specified offset indicate that the corresponding rcu_head
931  * structure can be handled by kvfree_rcu()?
932  */
933 #define __is_kvfree_rcu_offset(offset) ((offset) < 4096)
934 
935 /**
936  * kfree_rcu() - kfree an object after a grace period.
937  * @ptr: pointer to kfree for both single- and double-argument invocations.
938  * @rhf: the name of the struct rcu_head within the type of @ptr,
939  *       but only for double-argument invocations.
940  *
941  * Many rcu callbacks functions just call kfree() on the base structure.
942  * These functions are trivial, but their size adds up, and furthermore
943  * when they are used in a kernel module, that module must invoke the
944  * high-latency rcu_barrier() function at module-unload time.
945  *
946  * The kfree_rcu() function handles this issue.  Rather than encoding a
947  * function address in the embedded rcu_head structure, kfree_rcu() instead
948  * encodes the offset of the rcu_head structure within the base structure.
949  * Because the functions are not allowed in the low-order 4096 bytes of
950  * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
951  * If the offset is larger than 4095 bytes, a compile-time error will
952  * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can
953  * either fall back to use of call_rcu() or rearrange the structure to
954  * position the rcu_head structure into the first 4096 bytes.
955  *
956  * Note that the allowable offset might decrease in the future, for example,
957  * to allow something like kmem_cache_free_rcu().
958  *
959  * The BUILD_BUG_ON check must not involve any function calls, hence the
960  * checks are done in macros here.
961  */
962 #define kfree_rcu(ptr, rhf...) kvfree_rcu(ptr, ## rhf)
963 
964 /**
965  * kvfree_rcu() - kvfree an object after a grace period.
966  *
967  * This macro consists of one or two arguments and it is
968  * based on whether an object is head-less or not. If it
969  * has a head then a semantic stays the same as it used
970  * to be before:
971  *
972  *     kvfree_rcu(ptr, rhf);
973  *
974  * where @ptr is a pointer to kvfree(), @rhf is the name
975  * of the rcu_head structure within the type of @ptr.
976  *
977  * When it comes to head-less variant, only one argument
978  * is passed and that is just a pointer which has to be
979  * freed after a grace period. Therefore the semantic is
980  *
981  *     kvfree_rcu(ptr);
982  *
983  * where @ptr is the pointer to be freed by kvfree().
984  *
985  * Please note, head-less way of freeing is permitted to
986  * use from a context that has to follow might_sleep()
987  * annotation. Otherwise, please switch and embed the
988  * rcu_head structure within the type of @ptr.
989  */
990 #define kvfree_rcu(...) KVFREE_GET_MACRO(__VA_ARGS__,		\
991 	kvfree_rcu_arg_2, kvfree_rcu_arg_1)(__VA_ARGS__)
992 
993 #define KVFREE_GET_MACRO(_1, _2, NAME, ...) NAME
994 #define kvfree_rcu_arg_2(ptr, rhf)					\
995 do {									\
996 	typeof (ptr) ___p = (ptr);					\
997 									\
998 	if (___p) {									\
999 		BUILD_BUG_ON(!__is_kvfree_rcu_offset(offsetof(typeof(*(ptr)), rhf)));	\
1000 		kvfree_call_rcu(&((___p)->rhf), (rcu_callback_t)(unsigned long)		\
1001 			(offsetof(typeof(*(ptr)), rhf)));				\
1002 	}										\
1003 } while (0)
1004 
1005 #define kvfree_rcu_arg_1(ptr)					\
1006 do {								\
1007 	typeof(ptr) ___p = (ptr);				\
1008 								\
1009 	if (___p)						\
1010 		kvfree_call_rcu(NULL, (rcu_callback_t) (___p));	\
1011 } while (0)
1012 
1013 /*
1014  * Place this after a lock-acquisition primitive to guarantee that
1015  * an UNLOCK+LOCK pair acts as a full barrier.  This guarantee applies
1016  * if the UNLOCK and LOCK are executed by the same CPU or if the
1017  * UNLOCK and LOCK operate on the same lock variable.
1018  */
1019 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
1020 #define smp_mb__after_unlock_lock()	smp_mb()  /* Full ordering for lock. */
1021 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
1022 #define smp_mb__after_unlock_lock()	do { } while (0)
1023 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
1024 
1025 
1026 /* Has the specified rcu_head structure been handed to call_rcu()? */
1027 
1028 /**
1029  * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu()
1030  * @rhp: The rcu_head structure to initialize.
1031  *
1032  * If you intend to invoke rcu_head_after_call_rcu() to test whether a
1033  * given rcu_head structure has already been passed to call_rcu(), then
1034  * you must also invoke this rcu_head_init() function on it just after
1035  * allocating that structure.  Calls to this function must not race with
1036  * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation.
1037  */
1038 static inline void rcu_head_init(struct rcu_head *rhp)
1039 {
1040 	rhp->func = (rcu_callback_t)~0L;
1041 }
1042 
1043 /**
1044  * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()?
1045  * @rhp: The rcu_head structure to test.
1046  * @f: The function passed to call_rcu() along with @rhp.
1047  *
1048  * Returns @true if the @rhp has been passed to call_rcu() with @func,
1049  * and @false otherwise.  Emits a warning in any other case, including
1050  * the case where @rhp has already been invoked after a grace period.
1051  * Calls to this function must not race with callback invocation.  One way
1052  * to avoid such races is to enclose the call to rcu_head_after_call_rcu()
1053  * in an RCU read-side critical section that includes a read-side fetch
1054  * of the pointer to the structure containing @rhp.
1055  */
1056 static inline bool
1057 rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f)
1058 {
1059 	rcu_callback_t func = READ_ONCE(rhp->func);
1060 
1061 	if (func == f)
1062 		return true;
1063 	WARN_ON_ONCE(func != (rcu_callback_t)~0L);
1064 	return false;
1065 }
1066 
1067 /* kernel/ksysfs.c definitions */
1068 extern int rcu_expedited;
1069 extern int rcu_normal;
1070 
1071 #endif /* __LINUX_RCUPDATE_H */
1072